Loading Events

« All Events

Workshop YES: ""Optimal Transport, Statistics, Machine Learning and moving in between"

Sep 5 - Sep 9

Summary

Theory and Computational Aspects
In the last 30 years, the theory of Optimal Transport has emerged as a fertile field of inquiry, and a diverse tool for exploring applications within and beyond mathematics, in such diverse fields as economics, meteorology, geometry, fluid mechanics, design problems, theoretical chemistry and engineering. More recently, due to unexpected connections, as for instance with Data Sciences and Statistical Inference, theoretical and computational aspects of Optimal Transport regained substantial attention of both theorists and practitioners.
Optimal Transport principles have been applied very recently in formulating solutions to problems in the area of  statistical inference, and numerous machine learning problems such as generative learning, transfer learning,  distributionally robust optimization, and so on, with impressive results.
From a statistical point of view, transportation distances are very appealing since they quantify in a natural and meaningful way the notion of perturbations of a probability distribution (so are particularly suitable when considering mixture distributions). From a machine learning point of view, Wasserstein distances provide a very good way to compare two instances, by quantifying how much one needs to "warp" one instance to reach the other (for example, to measure similarities between images). Moreover, Optimal Transport methods have had a significant impact in the theory of nonlinear PDEs, in metric geometry, and in the development of functional inequalities.
Despite all these successes, Optimal Transport methods in Machine Learning and Statistical Inference have been not exhaustively explored, which makes discussion on the topic among experts in the field very timely. In fact, it would allow us to build new bridges between PDEs and Statistics via the discovery of new connections with Optimal Transport.

Goals of the workshop: The goals of this workshop are:
(1) facilitate and promote dialogue
and cross-fertilization between theoreticians/practitioners in the field of Statistics and Optimal Transport;
(2) provide a natural entry point for junior researchers to get acquainted with this
broad field of research.

In light of the the above and the unarguable interdisciplinary aspect of optimal transport research, the organizers believe the time is ripe to foster and promote the development of community that thrives at the intersections of statistics,  analysis and machine learning.

Programme Description:  Event consisting of two intertwined parts:
(I) School (days 1-3): during that period the program will consist on introductory and more advanced lectures given by experts from different fields, with the goal of equipping young researchers with basic-to-advanced mathematical tools in statistics and optimal transport.
(II) Workshop (days 3-5): a workshop format featuring talks given by a mix of young and senior researchers on recent developments at the interface of statistics and optimal transportation. Topics include (but are not limited to) statistical inference, concentration inequalities, Wasserstein barycenters and computational aspects.

For both parts we will ensure there is ample time for discussions and cross-fertilization. The final day of the school will be partially devoted to discussion groups focused on future perspectives and open problems proposed by the minicourse lecturers.
The overall event will (not exclusively) target a broad audience of early stage mathematicians (postdoctoral researchers, Ph.D. students and advanced master level students), as well as more senior and established researchers.
Confirmed mini-course speakers are:

Jonathan Niles-Weed NYU
Marco Cuturi CREST - ENSAE, Apple ML Research
Victor M. Panaretos EPFL

Registration

Registration is compulsory, the form will come online soon.

 

Further information to follow.

Details

Start:
Sep 5
End:
Sep 9
Event Categories:
,